The utility of artificial intelligence for DNA variant curation in the Mendelian genetic disorder familial...

The assessment of DNA sequence variants for biological function and their classification into risk categories is time-consuming. This has widened the gap between the rapid generation of genomic data and the issuing of clinical reports. The human input for variant curation requires the additional use of artificial intelligence (AI) to address efficiency and accuracy. A commercial AI tool called Franklin (Genoox Qiagen) was assessed for its usefulness in DNA variant curation. The testing model was the LDLR gene associated with familial hypercholesterolaemia (FH). A total of 327 curated LDLR variants available through ClinGen's Variant Curation Expert Panel (VCEP) were used to assess the utility of AI in variant curation. Franklin performed satisfactorily in the benign/likely benign DNA variant risk categories. All VCEP pathogenic DNA variants were classified correctly, although mixed with likely pathogenic cases. However, the latter could be identified by an experienced human variant curator. Franklin rated poorly in the likely pathogenic and variant of uncertain significance (VUS) DNA risk classes. Variant curations from Franklin were 100% reproducible. Although the version of Franklin tested will not replace the human variant curator, the study identified an assistantial role to reduce the turnaround time. It showed how an AI tool could be trained to perform better to enhance human-generated variant curations. With its speed, a tool such as Franklin will facilitate quality activities, training and review of results in both clinical and research laboratories.